In RL, the probabilities are a bit more within our grasp, but even then don't lend themselves to quick analysis. In addition, there are still a fair number of metrics that don't always agree completely.
That being said, determining the engine's probabilities is, IMHO, out of reach w/o a great deal of testing and collecting data. Frankly, your intuitive method probably approaches the best estimation of base-killing strengths in the outfield and catcher arm, with regard to steal chances. Best of luck, having really not helped you at all.
